NLPinAI 2021 Abstracts


Area 1 - NLPinAI

Full Papers
Paper Nr: 1
Title:

Strengthening Low-resource Neural Machine Translation through Joint Learning: The Case of Farsi-Spanish

Authors:

Benyamin Ahmadnia, Raul Aranovich and Bonnie J. Dorr

Abstract: This paper describes a systematic study of an approach to Farsi-Spanish low-resource Neural Machine Translation (NMT) that leverages monolingual data for joint learning of forward and backward translation models. As is standard for NMT systems, the training process begins using two pre-trained translation models that are iteratively updated by decreasing translation costs. In each iteration, either translation model is used to translate monolingual texts from one language to another, to generate synthetic datasets for the other translation model. Two new translation models are then learned from bilingual data along with the synthetic texts. The key distinguishing feature between our approach and standard NMT is an iterative learning process that improves the performance of both translation models, simultaneously producing a higher-quality synthetic training dataset upon each iteration. Our empirical results demonstrate that this approach outperforms baselines.

Paper Nr: 3
Title:

Augmented Spanish-Persian Neural Machine Translation

Authors:

Benyamin Ahmadnia and Raul Aranovich

Abstract: Neural Machine Translation (NMT) performs training of a neural network employing an encoder-decoder architecture. However, the quality of the neural-based translations predominantly depends on the availability of a large amount of bilingual training dataset. In this paper, we explore the performance of translations predicted by attention-based NMT systems for Spanish to Persian low-resource language pairs. We analyze the errors of NMT systems that occur in the Persian language and provide an in-depth comparison of the performance of the system based on variations in sentence length and size of the training dataset. We evaluate our translation results using BLEU and human evaluation measures based on the adequacy, fluency, and overall rating.

Paper Nr: 5
Title:

ArabiaNer: A System to Extract Named Entities from Arabic Content

Authors:

Mohammad Hudhud, Hamed Abdelhaq and Fadi Mohsen

Abstract: The extraction of named entities from unstructured text is a crucial component in numerous Natural Language Processing (NLP) applications such as information retrieval, question answering, machine translation, to name but a few. Named-entity Recognition (NER) aims at locating proper nouns from unstructured text and classifying them into a predefined set of types, such as persons, locations, and organizations. There has been extensive research on improving the accuracy of NER in English text. For other languages such as Arabic, extracting Named-entities is quite challenging due to its morphological structure. In this paper, we introduce ArabiaNer, a system employing Conditional Random Field (CRF) learning algorithm with extensive feature engineering steps to effectively extract Arabic named Entities. ArabiaNer produced state-of-the-art results with f1-score of 91.31% when applied on the ANERcrop dataset.

Paper Nr: 6
Title:

CNN-LSTM-CRF for Aspect-Based Sentiment Analysis: A Joint Method Applied to French Reviews

Authors:

Bamba Kane, Ali Jrad, Abderrahman Essebbar, Ophélie Guinaudeau, Valeria Chiesa, Ilhem Quénel and Stéphane Chau

Abstract: Aspect Based Sentiment Analysis (ABSA) aims to detect the different aspects addressed in a text and the sentiment associated to each of them. There exists a lot of work on this topic for the English language, but only few models are adapted for French ABSA. In this paper, we propose a new model for ABSA, named CLC, which combines CNN (Convolutional Neural Network), Bidirectional LSTM (Long Short-Term Memory) and CRF (Conditional Random Field). We demonstrate herein its great performance on the SemEval2016 French dataset. We prove that our CLC model outperforms the state-of-the-art models for French ABSA. We also prove that CLC is well adapted for other languages such as English. One main strength of CLC is its ability to detect the aspects and the associated sentiments in a joint manner, unlike the state-of-the-art models which detect them separately.

Paper Nr: 11
Title:

Formal Validation for Natural Language Programming using Hierarchical Finite State Automata

Authors:

Yue Zhan and Michael S. Hsiao

Abstract: Natural language programming (NLPr) is a sub-field of natural language processing (NLP) that provides a bridge between natural languages (NL) and programming languages (PL), allowing users to design programs in the form of structured NL documents. Due to the imprecise and ambiguous nature of NL, it is essential to ensure the correctness of translation for critical applications where errors are unacceptable. Machine learning-based approaches for error checking are insufficient as it can be difficult for even the most sophisticated models to capture all the relevant intricacies of a natural language. Automata offer a formalism that has been used in compiling programming languages, and this paper extends automata-based methods to validating programs written in natural languages. In particular, we propose a hierarchically structured finite-state automaton, modeled based on domain-specific knowledge, for NLPr input validation and semantic error reporting. Experimental results from validating a set of collected NL sentences show that the proposed validation and error reporting can catch the unexpected input components while validating the semantics.

Short Papers
Paper Nr: 7
Title:

Aspect Based Sentiment Analysis using French Pre-Trained Models

Authors:

Abderrahman Essebbar, Bamba Kane, Ophélie Guinaudeau, Valeria Chiesa, Ilhem Quénel and Stéphane Chau

Abstract: Aspect Based Sentiment Analysis (ABSA) is a fine-grained task compared to Sentiment Analysis (SA). It aims to detect each aspect evoked in a text and the sentiment associated to each of them. For English language, many works using Pre-Trained Models (PTM) exits and many annotated open datasets are also available. For French Language, many works exits in SA and few ones for ABSA. We focus on aspect target sentiment analysis and we propose an ABSA using French PTM like multilingual BERT (mBERT), CamemBERT and FlauBERT. Three different fine-tuning methods: Fully-Connected, Sentences Pair Classification and Attention Encoder Network, are considered. Using the SemEval2016 French reviews datasets for ABSA, our fine-tuning models outperforms the state-of-the-art French ABSA methods and is robust for the Out-Of-Domain dataset.

Paper Nr: 8
Title:

Neural Machine Translation for Amharic-English Translation

Authors:

Andargachew M. Gezmu, Andreas Nürnberger and Tesfaye B. Bati

Abstract: This paper describes neural machine translation between orthographically and morphologically divergent languages. Amharic has a rich morphology; it uses the syllabic Ethiopic script. We used a new transliteration technique for Amharic to facilitate vocabulary sharing. To tackle the highly inflectional morphology and to make an open vocabulary translation, we used subwords. Furthermore, the research was conducted on low-data conditions. We used the transformer-based neural machine translation architecture by tuning the hyperparameters for low-data conditions. In the automatic evaluation of the strong baseline, word-based, and subword-based models trained on a public benchmark dataset, the best subword-based models outperform the baseline models by approximately six up to seven BLEU.

Paper Nr: 9
Title:

Memory State Tracker: A Memory Network based Dialogue State Tracker

Authors:

Di Wang and Simon O’Keefe

Abstract: Dialogue State Tracking (DST) is a core component towards task oriented dialogue system. It fills manually-set slots at each turn of an utterance, which indicate the current topics or user requirement. In this work we propose a memory based state tracker that includes a memory encoder which encodes the dialogue history into a memory vector, and then connects to a pointer network which makes predictions. Our model reached a joint goal accuracy of 49.16% on MultiWOZ 2.0 data set (Budzianowski et al., 2018) and 47.27% on MultiWOZ 2.1 data set (Eric et al., 2019), outperforming the benchmark result.